Letter https://doi.org/10.

1038/s41586-019-1043-4

Complex societies precede moralizing gods

throughout world history
Harvey Whitehouse1,15, Pieter François1,2,15, Patrick E. Savage1,3,15*, Thomas E. Currie4, Kevin C. Feeney5, Enrico Cioni6,
Rosalind Purcell6, Robert M. Ross1,7,8, Jennifer Larson9, John Baines10, Barend ter Haar11, Alan Covey12 & Peter Turchin13,14

The origins of religion and of complex societies represent
evolutionary puzzles1–8. The ‘moralizing gods’ hypothesis offers
a solution to both puzzles by proposing that belief in morally
concerned supernatural agents culturally evolved to facilitate
cooperation among strangers in large-scale societies9–13. Although
previous research has suggested an association between the presence
of moralizing gods and social complexity3,6,7,9–18, the relationship
between the two is disputed9–13,19–24, and attempts to establish
causality have been hampered by limitations in the availability of
detailed global longitudinal data. To overcome these limitations,
here we systematically coded records from 414 societies that span
the past 10,000 years from 30 regions around the world, using
51 measures of social complexity and 4 measures of supernatural
enforcement of morality. Our analyses not only confirm the
association between moralizing gods and social complexity, but
also reveal that moralizing gods follow—rather than precede—
large increases in social complexity. Contrary to previous
predictions9,12,16,18, powerful moralizing ‘big gods’ and prosocial
supernatural punishment tend to appear only after the emergence
of ‘megasocieties’ with populations of more than around one million
people. Moralizing gods are not a prerequisite for the evolution
of social complexity, but they may help to sustain and expand
complex multi-ethnic empires after they have become established.
By contrast, rituals that facilitate the standardization of religious
traditions across large populations25,26 generally precede the
appearance of moralizing gods. This suggests that ritual practices
were more important than the particular content of religious belief
to the initial rise of social complexity.
Supernatural agents that punish direct affronts to themselves (for
example, failure to perform sacrifices or observe taboos) are commonly
represented in global history, but rarely are such deities believed to
punish moral violations in interactions between humans2. Recent mil-
lennia, however, have seen the rise and spread of several ‘prosocial
religions’, which include either powerful ‘moralizing high gods’ (MHG;
for example, the Abrahamic God) or more general ‘broad supernatu-
ral punishment’ (BSP) of moral transgressions (for example, karma in
Buddhism)9,12,16–18. Such moralizing gods may have provided a crucial
mechanism for overcoming the classic free-rider problem in large-scale
societies11. The association between moralizing gods and complex
societies has been supported by two forms of evidence: psychological
experiments3,6,27,28 and cross-cultural comparative analyses7,11,14–18,20.
The contributions of theistic beliefs to cooperation, as well as the his-
torical question of whether moralizing gods precede or follow the estab-
lishment of large-scale cooperation, have been much debated9,10,12,23,24.
Three recent studies that explicitly model temporal causality have come
to contrasting conclusions. One study, which applied phylogenetic
comparative methods to infer historical changes in Austronesian reli-
gions, reported that moralizing gods (BSP but not MHG) preceded the
evolution of complex societies16. The same conclusion was reached
in an analysis of historical and archaeological data from Viking-age
Scandinavia18. By contrast, another study of Eurasian empires has
reported that moralizing gods followed—rather than preceded—the
rise of complex, affluent societies20. However, all of these studies are
restricted in geographical scope and use proxies for social complexity
that the authors themselves concede are ‘very crude’20 (for example,
the binary classification of societies as of either high or low complexity).
To overcome these limitations, we used ‘Seshat: Global History
Databank’29, a repository of standardized data on social structure, reli-
gion and other domains for hundreds of societies throughout world his-
tory. In contrast to other databases that attempt to model history using
contemporary ethnographic data, Seshat directly samples over time as
well as space. Seshat also includes estimates of expert disagreement and
uncertainty, and uses more-detailed variables than many databases.
To test the moralizing gods hypothesis, we coded data on 55 var-
iables from 414 polities (independent political units) that occupied
30 geographical regions from the beginning of the Neolithic period
to the beginning of Industrial and/or colonial periods (Fig. 1 and
Supplementary Data). We used a recently developed and validated
measure of social complexity that condenses 51 social complexity
variables (Extended Data Table 5) into a single principal component
that captures three quarters of the observed variation, which we call
‘social complexity’8. The remaining four variables were selected to test
the MHG and BSP subtypes of the moralizing gods hypothesis. The
MHG variable was coded following the MHG variable used as stand-
ard in the literature on this topic11,14–17,30, which requires that a high
god who created and/or governs the cosmos actively enforces human
morality. Because the concept of morality is complex, multidimensional
and in some respects culturally relative—and because not all moralizing
gods are ‘high gods’—we also coded three different variables related to
BSP that are specifically relevant to prosocial cooperation: reciprocity,
fairness and in-group loyalty. For analysis, these three variables were
combined into a single BSP variable. The Methods, Supplementary
Information and http://seshatdatabank.info/methods/codebook pro-
vide further methodological details, definitions and justifications,
including a discussion of the relationship between MHG, BSP and big
gods.
Figure 1 and Extended Data Table 1 show the temporal and geo-
graphical distribution of the appearance of moralizing gods in our
sample. Although societies in all 30 regions possessed beliefs about
appeasing supernatural agents through the performance of rituals, in 10
out of the 30 regions, there was no evidence for moralizing gods before
their introduction under colonial powers. The remaining 20 regions

1
Centre for the Study of Social Cohesion, University of Oxford, Oxford, UK. 2St Benet’s Hall, Oxford, UK. 3Faculty of Environment and Information Studies, Keio University, Fujisawa, Japan.
4
Human Behaviour & Cultural Evolution Group, Department of Biosciences, University of Exeter, Penryn, UK. 5School of Computer Science and Statistics, Trinity College Dublin, Dublin, Ireland.
6
Seshat: Global History Databank, Evolution Institute, San Antonio, FL, USA. 7ARC Centre of Excellence in Cognition and its Disorders and Department of Psychology, Royal Holloway, University of
London, Egham, UK. 8Department of Anthropology and Archaeology, University of Bristol, Bristol, UK. 9Department of Modern & Classical Language Studies, Kent State University, Kent, OH, USA.
10
Faculty of Oriental Studies, University of Oxford, Oxford, UK. 11Department of Chinese Language and Culture, Asia-Africa-Institute, University of Hamburg, Hamburg, Germany. 12Department of
Anthropology, University of Texas at Austin, Austin, TX, USA. 13Department of Ecology and Evolutionary Biology, University of Connecticut, Storrs, CT, USA. 14Complexity Science Hub Vienna, Wien,
Austria. 15These authors contributed equally: Harvey Whitehouse, Pieter François, Patrick E. Savage. *e-mail: psavage@sfc.keio.ac.jp

N A t U r e | www.nature.com/nature
 RESEARCH Letter
1.0
2.1
0.2 0.9
Earliest precolonial evidence
of moralizing gods (ka) 0.5
Zoroastrianism
Abrahamic
Other MHG
Buddhism
Other BSP
Absent
Fig. 1 | Locations of the 30 sampled regions on the world map, labelled
according to precolonial evidence of moralizing gods. The area of
each circle is proportional to social complexity of the earliest polity with
moralizing gods to occupy the region or the latest precolonial polity for
regions without precolonial moralizing gods. For regions with precolonial
moralizing gods, the date of earliest evidence of such beliefs is displayed in
thousands of years ago (ka), coloured by type of moralizing gods.
displayed a diverse range of 15 different systems of belief in moraliz-
ing gods: in some, the first evidence of moralizing gods came in the
form of MHG and in others it came in the form of BSP (Extended
Data Table 1). The first appearance of moralizing gods in our sample
was in Egypt, where the concept of supernatural enforcement of
Maat (order) is attested by the Second Dynasty, around 2800 bc. This
was followed by sporadic appearances in local religions throughout
Eurasia (Mesopotamia (around 2200 bc), Anatolia (around 1500 bc)
and China (around 1000 bc)) before the wider spread of transnational
religions began during the first millennium bc with Zoroastrianism
and Buddhism, followed later by Christianity and Islam. Although
Christianity and Islam would eventually become the most widespread
religions, local forms of moralizing gods were present well before they
arrived in most regions (for example, Roman gods were believed to
punish oath-breaking from as early as 500 bc, almost a millennium
before Christianity was adopted as the official Roman religion). The
diverse range of religious systems represented in our global sample
makes it possible to draw more general conclusions about religion than
have previously been possible.
Although our sampling scheme reduces non-independence, our
polities still cannot be considered statistically independent because
of the historical relationships among them. We controlled for these
using a logistic regression model to account for temporal, geographi-
cal and cultural dependencies in the global distribution of moralizing
gods (see Methods). This analysis revealed that social complexity was
a stronger predictor of moralizing gods than temporal, geographical or
linguistic relationships, and remained highly significant even after con-
trolling for these relationships (z = 6.8, degrees of freedom (d.f.) = 800,
P < 1 × 10−11; Extended Data Table 2), conceptually replicating pre-
vious studies7,11,14,15.
The moralizing gods hypothesis posits a ‘statistical causal relation-
ship’10 in which moralizing gods facilitate the evolution of complex
societies9,12,16–18. This indicates that, on average, social complexity
should increase more rapidly following the appearance of moralizing
gods. To test this prediction, we conducted time-series analyses of the
12 regions for which social complexity data were available both before
and after the appearance of moralizing gods (Fig. 2, Extended Data
N A t U r e | www.nature.com/nature
1.7
2.5 2.5
3.5
3 1.4
4.2
2.5
4.8
1.6 2.3 1.7 0.2
0.3
1.2
The three transnational religious systems that represent the first
appearance of moralizing gods in more than one region—Zoroastrianism,
Abrahamic religions (Judaism, Islam and Christianity) and Buddhism—are
coloured red, orange and blue, respectively, whereas other local religious
systems with beliefs in MHG or BSP are coloured yellow and purple,
respectively. See Extended Data Table 1 for further details.
Table 1 and Extended Data Fig. 1). Notably, average rates of increase
of social complexity were over five times greater before—not after—
the appearance of moralizing gods (paired t-test, t = −6.6, d.f. = 199,
P < 1 × 10−9; Fig. 2). This trend was significant both globally and
individually for 10 out of the 12 regional time-series analyses (Extended
Data Table 1 and Extended Data Fig. 1). None of these 12 regions dis-
played a significantly greater rate of increase in social complexity after
the appearance of moralizing gods than before. Robustness analyses
showed that our primary finding of higher rates of increasing social
complexity before the appearance of moralizing gods was present
regardless of the type of moralizing gods (MHG or BSP), the choice of
variables used to estimate social complexity, uncertainty in the timing
of appearance of moralizing gods, or the time windows used to estimate
rates of change in social complexity (Extended Data Table 4).
In summary, although our analyses are consistent with previous stud-
ies that show an association between moralizing gods and complex
societies7,11,14–18,30, we find that moralizing gods usually follow—rather
than precede—the rise of social complexity. Notably, most societies that
exceeded a certain social complexity threshold developed a conception
of moralizing gods. Specifically, in 10 out of the 12 regions analysed,
the transition to moralizing gods came within 100 years of exceeding a
social complexity value of 0.6 (which we call a megasociety, as it corre-
sponds roughly to a population in the order of one million; Extended
Data Fig. 1). This megasociety threshold does not seem to correspond
to the point at which societies develop writing, which might have sug-
gested that moralizing gods were present earlier but were not preserved
archaeologically. Although we cannot rule out this possibility, the fact
that written records preceded the development of moralizing gods in
9 out of the 12 regions analysed (by an average period of 400 years;
Supplementary Table 2)—combined with the fact that evidence for
moralizing gods is lacking in the majority of non-literate societies2—
suggests that such beliefs were not widespread before the invention
of writing. The few small-scale societies that did display precolonial
evidence of moralizing gods came from regions that had previously
been used to support the claim that moralizing gods contributed to the
rise of social complexity (Austronesia16 and Iceland18), which suggests
that such regions are the exception rather than the rule.

a 1.0
0.8
Social complexity
0.6
0.4
Moralizing gods
0.2
Doctrinal rituals
0 0
–2,000 –1,000 0 1,000
Time
(years before/after moralizing gods)
Fig. 2 | Social complexity before and after the appearance of
moralizing gods. a, Time series showing mean social complexity over
time for 2,000 years before and after the appearance of moralizing gods.
n = 12 regions with social complexity data for before and after moralizing
gods. Social complexity has been scaled so that the society with the highest
social complexity (Qing Dynasty, China, around ad 1900) has a value
of 1 and the society with the lowest social complexity (Early Woodland,
Illinois, USA, around 400 bc) has a value of 0. Vertical bands represent
the period in which moralizing gods and doctrinal rituals first appeared.
All errors represent 95% confidence intervals, with the exception of the
vertical bar for moralizing gods, which represents the mean duration of the
Conversely, of the societies in the ten regions that did not develop
precolonial moralizing gods, only one exceeded the megasociety
threshold (the short-lived Inca Empire, social complexity = 0.61).
This suggests that, even if moralizing gods do not cause the evolution
of complex societies, they may represent a cultural adaptation that is
necessary to maintain cooperation in such societies once they have
exceeded a certain size, perhaps owing to the need to subject diverse
populations in multi-ethnic empires to a common higher-level power9.
This may explain why moralizing gods spread when large empires con-
quer smaller—but still complex—societies (for example, the Spanish
conquest of the Incas). In some cases, moralizing doctrines may have
helped to stabilize empires, while also limiting further expansion; for
example, when emperor Ashoka adopted Buddhism and renounced
war following his final conquest of the Kalinga Kingdom that estab-
lished the maximum extent of the Mauryan empire.
Although our results do not support the view that moralizing gods
were necessary for the rise of complex societies, they also do not
support a leading alternative hypothesis that moralizing gods only
emerged as a byproduct of a sudden increase in affluence during a
first millennium bc ‘Axial Age’19–22. Instead, in three of our regions
(Egypt, Mesopotamia and Anatolia), moralizing gods appeared before
1500 bc. We propose that the standardization of beliefs and practices
via high-frequency repetition and enforcement by religious authorities
enabled the unification of large populations for the first time, establish-
ing common identities across states and empires25,26. Our data show
that doctrinal rituals standardized by routinization (that is, those per-
formed weekly or daily) or institutionalized policing (religions with
multiple hierarchical levels) significantly predate moralizing gods, by an
average of 1,100 years (t = 2.8, d.f. = 11, P = 0.018; Fig. 2a). Doctrinal
rituals precede moralizing gods in 9 out of the 12 regions analysed,
and even precede written records in 6 of these cases (by as much as
4,000 years in the case of Çatalhöyük in Anatolia; see Supplementary
Table 2). Although analyses of rates of change of social complex-
ity before and after the appearance of doctrinal rituals do not offer
conclusive support for the hypothesis that doctrinal rituals facilitate
increasing social complexity (Extended Data Table 3), these data do at
least suggest that doctrinal rituals led to the establishment of large-scale
religious identities. In the future, higher-quality and higher-resolution
Letter RESEARCH
b
80
60
Frequency
40
20
2,000 –15 –10 –5 0 5
Rates of change in
social complexity (SC per kyr)
polity in which moralizing gods appeared (because times are normalized
to the time of first evidence of moralizing gods, and there is thus no
variance in this parameter). b, Histogram of the differences in rates of
change in social complexity (SC) after minus before the appearance
of moralizing gods. n = 200 time windows from the 12 regions. kyr,
thousand years. The y axis represents the number of time windows out
of 200. See Extended Data Fig. 1 for data for each of the 12 regions and
Extended Data Fig. 2 for a version extending beyond 2,000 years before
and after moralizing gods. The analyses in this figure treat the presence of
either MHG or BSP as ‘moralizing gods’—see Extended Data Fig. 3 for an
alternative analysis restricted only to the presence of MHG.
archaeological data may allow for a more nuanced understanding of
the timing and possible coevolution of the rise of doctrinal rituals and
moralizing gods. Such data appear unlikely to affect our primary claim
that complex societies preceded moralizing gods, but this is an empiri-
cal question open to future testing.
We demonstrate how quantifying cultural characteristics of past
societies can contribute to longstanding debates about the evolution
of social complexity. Our results suggest that belief in moralizing gods
was not the only or even the main factor that enabled the expansion
of human societies, but may have occurred along with other features
of ritual practices and religion to facilitate cooperation in increasingly
complex social systems. In particular, an increase in ritual frequency
and doctrinal control may have facilitated the establishment of large-
scale collective identities before the spread of beliefs in moralizing
gods. Thus, when it comes to the initial rise of social complexity, how
you worship may ultimately have been more important than who you
worship.
Online content
Any methods, additional references, Nature Research reporting summaries, source
data, statements of data availability and associated accession codes are available at
https://doi.org/10.1038/s41586-019-1043-4.
Received: 29 January 2018; Accepted: 19 February 2019;
Published online xx xx xxxx.
1. Darwin, C. The Descent of Man, and Selection in Relation to Sex (John Murray,
London, 1871).
2. Bellah, R. N. Religion in Human Evolution: From the Paleolithic to the Axial Age
(Harvard Univ. Press, Cambridge, 2011).
3. Purzycki, B. G. et al. Moralistic gods, supernatural punishment and the
expansion of human sociality. Nature 530, 327–330 (2016).
4. Watts, J., Sheehan, O., Atkinson, Q. D., Bulbulia, J. & Gray, R. D. Ritual human
sacrifice promoted and sustained the evolution of stratified societies. Nature
532, 228–231 (2016).
5. Currie, T. E., Greenhill, S. J., Gray, R. D., Hasegawa, T. & Mace, R. Rise and fall of
political complexity in island South-East Asia and the Pacific. Nature 467,
801–804 (2010).
6. Henrich, J. et al. Markets, religion, community size, and the evolution of fairness
and punishment. Science 327, 1480–1484 (2010).
7. Botero, C. A. et al. The ecology of religious beliefs. Proc. Natl Acad. Sci. USA 111,
16784–16789 (2014).
N A t U r e | www.nature.com/nature
 RESEARCH Letter
8. Turchin, P. et al. Quantitative historical analysis uncovers a single dimension
of complexity that structures global variation in human social organization.
Proc. Natl Acad. Sci. USA 115, E144–E151 (2018).
9. Norenzayan, A. et al. The cultural evolution of prosocial religions. Behav. Brain
Sci. 39, e1 (2016).
10. Norenzayan, A. Big questions about big gods: response and discussion. Religion
Brain Behav. 5, 327–342 (2015).
11. Johnson, D. D. P. God’s punishment and public goods: a test of the
supernatural punishment hypothesis in 186 world cultures. Hum. Nat. 16,
410–446 (2005).
12. Johnson, D. D. P. The wrath of the academics: criticisms, applications, and
extensions of the supernatural punishment hypothesis. Religion Brain Behav. 8,
320–350 (2018).
13. Schloss, J. P. & Murray, M. J. Evolutionary accounts of belief in supernatural
punishment: a critical review. Religion Brain Behav. 1, 46–99 (2011).
14. Roes, F. L. & Raymond, M. Belief in moralizing gods. Evol. Hum. Behav. 24,
126–135 (2003).
15. Peoples, H. C. & Marlowe, F. W. Subsistence and the evolution of religion.
Hum. Nat. 23, 253–269 (2012).
16. Watts, J. et al. Broad supernatural punishment but not moralizing high gods
precede the evolution of political complexity in Austronesia. Proc. R. Soc. B 282,
20142556 (2015).
17. Gray, R. D. & Watts, J. Cultural macroevolution matters. Proc. Natl Acad. Sci. USA
114, 7846–7852 (2017).
18. Raffield, B., Price, N. & Collard, M. Religious belief and cooperation: a view from
Viking-age Scandinavia. Religion Brain Behav. 9, 2–22 (2019).
19. Baumard, N. & Boyer, P. Explaining moral religions. Trends Cogn. Sci. 17,
272–280 (2013).
20. Baumard, N., Hyafil, A., Morris, I. & Boyer, P. Increased affluence explains the
emergence of ascetic wisdoms and moralizing religions. Curr. Biol. 25, 10–15
(2015).
21. Mullins, D. A. et al. A systematic assessment of ‘Axial Age’ proposals
using global comparative historical evidence. Am. Sociol. Rev. 83, 596–626
(2018).
22. Purzycki, B. G. et al. Material security, life history, and moralistic religions: a
cross-cultural examination. PLoS ONE 13, e0193856 (2018).
23. Stausberg, M. Big gods in review: introducing Ara Norenzayan and his critics.
Religion 44, 592–608 (2014).
24. McKay, R. & Whitehouse, H. Religion and morality. Psychol. Bull. 141, 447–473
(2015).
25. Whitehouse, H. Modes of Religiosity: A Cognitive Theory of Religious Transmission
(AltaMira, Walnut Creek, 2004).
26. Whitehouse, H., François, P. & Turchin, P. The role of ritual in the evolution of
social complexity: five predictions and a drum roll. Cliodynamics 6, 199–216
(2015).
27. Norenzayan, A. & Shariff, A. F. The origin and evolution of religious prosociality.
Science 322, 58–62 (2008).
28. Shariff, A. F., Willard, A. K., Andersen, T. & Norenzayan, A. Religious priming: a
meta-analysis with a focus on prosociality. Pers. Soc. Psychol. Rev. 20, 27–48
(2016).
N A t U r e | www.nature.com/nature
29. Turchin, P. et al. Seshat: The Global History Databank. Cliodynamics 6, 77–107
(2015).
30. Murdock, G. P. Ethnographic atlas: a summary. Ethnology 6, 109–236 (1967).
Acknowledgements We thank Q. Atkinson and A. Willard for feedback on
an earlier version of the manuscript and E. Postma for discussions on parts
of the statistical analyses. We acknowledge the contributions of our team of
research assistants, post-doctoral researchers, consultants and experts. See
http://www.seshatdatabank.info for a comprehensive list of private donors,
partners, experts and consultants. This work was supported by an ESRC Large
Grant entitled ‘Ritual, Community, and Conflict’ (REF RES-060-25-0085), a
John Templeton Foundation grant to the Evolution Institute entitled ‘Axial-Age
Religions and the Z-Curve of Human Egalitarianism’, a Tricoastal Foundation
grant to the Evolution Institute entitled ‘The Deep Roots of the Modern World:
The Cultural Evolution of Economic Growth and Political Stability’, an Advanced
Grant (‘Ritual Modes: Divergent modes of ritual, social cohesion, prosociality,
and conflict’, grant agreement no. 694986) and a Starter Grant (‘The Cultural
Evolution & Ecology of Institutions’, grant agreement no. 716212) from the
European Research Council (ERC) under the European Union’s Horizon 2020
Research and Innovation Programme, an award from the Templeton World
Charity Foundation entitled ‘Cognitive and Cultural Foundations of Religion
and Morality’ (TWCF0164), a Keio Research Institute at SFC Startup Grant, a
Keio Gijuku Academic Development Fund Individual Grant and a grant from
the European Union Horizon 2020 Research and Innovation Programme
(grant agreement no. 644055 (ALIGNED, www.aligned-project.eu)).
Reviewer information Nature thanks Carol Ember, Peter J. Richerson and the
other anonymous reviewer(s) for their contribution to the peer review of this
work.
Author contributions H.W., P.F. and P.E.S. designed the study, with input from
P.T., T.E.C., K.C.F. and R.M.R.; E.C. and R.P. coded the religion and ritual data, with
additional input from J.L., J.B., B.t.H and A.C.; P.E.S. analysed the data, with input
from H.W., P.F., P.T., T.E.C. and R.M.R.; P.E.S., H.W. and P.F. drafted the manuscript.
Competing interests The authors declare no competing interests.
Additional information
Extended data is available for this paper at https://doi.org/10.1038/s41586-
019-1043-4.
Supplementary information is available for this paper at https://doi.org/
10.1038/s41586-019-1043-4.
Reprints and permissions information is available at http://www.nature.com/
reprints.
Correspondence and requests for materials should be addressed to P.E.S.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
© The Author(s), under exclusive licence to Springer Nature Limited 2019

Methods
Seshat: Global History Databank overview. Seshat (http://seshatdatabank.
info/) is a vast database of information about global history from the end of the
Palaeolithic period up to the Industrial Revolution. Eventually, it is intended that
Seshat will cover the history of all past human societies; however, initially the goal
has been to capture as much diversity in global history as possible. We therefore
created a stratified sample of past societies by identifying ten world regions dis-
tributed as widely as possible across the Earth’s surface and within each of those
regions designated three ‘natural geographical areas’ (NGAs) with discrete eco-
logical boundaries, on average about 10,000 km2 in size, thus creating an initial
sampling scheme of 30 such areas around the world. To maximize diversity in
the sample, for each world region we chose one NGA in which social complexity
emerged early (for example, Egypt), one in which it arose relatively recently (for
example, Iceland) and one in which it emerged somewhere in the middle of the
range (for example, Japan)29. The 30 regions and their selection rationale have
previously been published29, before the start of data collection. Our aim was to
maximize variability in our global sample while minimizing historical relationships
between cultures.
Data on political systems (polities) that emerged and persisted in each of the
NGAs were then gathered and entered into Seshat in a continuous time series at
100-year intervals, going back as far into the history of that area as scholarly liter-
ature would allow (up to a maximum of roughly 10,000 years before present). In
the case of NGAs that contained clusters of very small-scale polities that share a
similar culture but are not under a single system of jurisdictional control, we refer
to these as ‘quasi-polities’ and code information on all of them generically, unless
information is available that would allow us to differentiate between these polities.
All variables for which data have been gathered and entered into Seshat are
derived from a Seshat Codebook that can be accessed and downloaded (http://
seshatdatabank.info/methods/codebook). The Codebook was designed by, and
is continually updated and extended in consultation with, a large network of pro-
fessional historians, archaeologists, anthropologists and other specialists whom
we refer to as ‘Seshat experts’. Most variables in Seshat require the data to take the
form of a number or numerical range or they specify a feature that can be coded
as absent, present or unknown (additionally coding items as ‘inferred present’ or
‘inferred absent’, where the evidence permits). The first step in data entry was for
trained research assistants to gather and input easily acquired data, and at the same
time to compile lists of data that are more difficult to interpret and that require
input from Seshat experts. Especially during the early phases of data entry, varia-
bles in the codebook were revised and improved through continuous discussions
between research assistants and Seshat experts. All data are linked to scholarly
sources, including peer-reviewed publications and personal communications from
established authorities. On occasions when Seshat experts disagree on a particular
coding, we kept a record of disagreements so that analyses could be run taking into
account contrasting interpretations. Once used for the purposes of data analysis
and publication, that version of the dataset was ‘frozen’ so that it could be inspected
by others and used for the purposes of replication. Nevertheless, the data in Seshat
continually evolves, as new sources are discovered and as new Seshat experts con-
tribute additional layers of interpretation.
The data analysed in the present paper focused on those sectors of the Seshat
Codebook concerned with social complexity, religion and ritual. A full account
of the social complexity variables has previously been published8, using 51 varia-
bles (Extended Data Table 5) associated with population size, hierarchy, territory,
governance, bureaucracy, infrastructure, record keeping, economic development
and other domains that were previously identified as potentially relevant meas-
ures of social complexity. This required engagement with a wide-ranging body
of literature on social complexity. Because previous researchers disagreed about
which dimensions of social complexity were the most important to emphasize (for
example, number of jurisdictional levels versus more-horizontal forms of com-
plexity; autocracy versus democracy; diversity of specialist roles versus centralized
coordination), we included proxies for all potentially relevant measures of social
complexity that had been identified in the literature. This inclusive strategy was
designed to allow us to investigate whether these different characteristics exhibited
strong relationships with each other and whether a single principal component cap-
tured most of the observed variation. Our analyses confirmed that both are indeed
the case. Furthermore, we found that different characteristics of social complexity
were highly predictable across different world regions8.
Whereas previous research has proposed an association between the rise of mor-
alizing gods and the evolution of social complexity, measures used in the past to
capture the latter have been comparatively crude. Variable selection and inclusion
for moralizing gods was informed by existing literature on so-called big gods, MHG
and BSP, as well as psychological and cross-cultural comparative research on the
hypothesized link between belief in moralizing gods and large-scale cooperation.
Data collection for the religion and ritual variables involved matching each fully
trained research assistant with one or more Seshat experts. Seshat experts provided
Letter RESEARCH
guidance on how to delineate the temporal and geographical boundaries of the
polity, assembled an initial reading list and—where necessary—helped to interpret
some of the key historiographical debates associated with the variables. Research
assistants then populated the variables with data and presented this to the Seshat
experts for review. The comments and suggestions made by the experts were then
implemented by the research assistants. The next stage required a second team of
fully trained research assistants to go over the gathered data and to conduct a series
of quality checks, including vetting of the footnotes and the use of correct syntax
for the machine-readable part of the data. Finally, this checked dataset was given
to the Seshat experts for review. The coding of religion and ritual data required
the input of experts every step of the way, given the frequent need for complex
and nuanced interpretation of the evidence. By contrast, the data required for the
social complexity variables frequently consisted of facts that research assistants
could procure with less supervision, allowing expert input and review to occur at
a later stage of the process.
Data coding. Social complexity. The 51 variables used to construct the overall social
complexity measure are shown in Extended Data Table 5. These variables were
chosen because they reflect common features associated with social complexity
and were grouped into nine complexity characteristics (polity population size,
capital population size, polity territory size, hierarchy, infrastructure, government,
information systems, texts and money). Details of coding definitions for these
variables have previously been published8,29.
MHG. For consistency with previous studies that have generally used the MHG
variable from the Ethnographic Atlas30, the presence of MHG was coded as a binary
variable on the basis of the original definition of a high god as “a spiritual being
who is believed to have created all reality and/or to be its ultimate governor, even
though his sole act was to create other spirits who, in turn, created or control the
natural world”31. The following categories were used: (1) absent or not reported,
(2) present but not active in human affairs, (3) present and active in human affairs
but not supportive of human morality and (4) present, active and specifically sup-
portive of human morality11,32.Thus, a coding of high gods as present, active and
specifically supportive of human morality was coded as a MHG being present,
whereas all other types were coded as absent.
BSP. The terms big gods and MHG are sometimes used interchangeably17, but
can have different connotations. The term MHG was previously developed and
defined, and it was proposed that high gods were associated with social complex-
ity (regardless of their moral concern)31. This definition of the MHG variable
was incorporated into the Ethnographic Atlas30, resulting in it being widely used
in cross-cultural research. These ideas were subsequently extended in the super-
natural punishment hypothesis11,12,33, in which the focus lay on the mechanism
of morality enforcement rather than high gods; however, the MHG variable was
used for testing because of the availability of previous research using this defini-
tion. The ideas have been further developed9,10,27,34 to include various additional
mechanisms, most notably including cultural group selection to explain the rapid
spread of moralizing gods without accompanying genetic changes.
The term big gods (defined as “powerful, omniscient, interventionist, morally
concerned gods”34) was originally the title of a monograph describing this theory.
Later, however, this definition was relaxed and it was emphasized that the term
big gods was a rhetorical device intended to include a broad range of morally
concerned supernatural agents, not only MHGs: “…powerful, all-knowing and
morally concerned supernatural agents who are believed to monitor social inter-
actions and to reward and sanction behaviours in ways that contribute to the cul-
tural success of the group, including practices that effectively transmit the faith.
Rhetorically, we call these ‘Big Gods’; however, we alert readers that we are referring
to a multidimensional continuum of supernatural agents in which big gods occupy
a particular corner of the space”9.
Subsequently an additional variable was developed that was called broad super-
natural punishment16, which arguably more closely matches this relaxed definition
of big gods than does the traditional MHG variable. BSP was defined as follows:
“For BSP to be coded as present in a culture there must be the concept of a super-
natural agent or process that reliably monitors and punishes selfish actions, and
this concept must (i) be widely advocated within the community, (ii) involve pun-
ishment of a broad range of selfish behaviours and (iii) apply to a wide range of
community members.”16.
Because selfish actions can occur in a variety of domains, Seshat subdivides the
types of supernatural enforcement of morality based on nine proposed categories
of morality35,36. For this study, we focused on three domains that are relevant to the
establishment of large-scale cooperation: (1) fairness (sharing of resources, such as
dividing disputed resources, bargaining or redistribution of wealth); (2) reciproc-
ity (for example, fulfilling contracts, returning gifts, repaying debts or upholding
trust); and (3) in-group loyalty (the need to remain loyal to unrelated members of
the same group; for example, helping coreligionists or going to war for one’s group).
BSP was coded as present if at least one of these three sub-types of selfish actions
was supernaturally enforced.
 RESEARCH Letter
Our robustness analyses, which evaluated BSP and MHG separately (Extended
Data Table 4) suggest that—rather than moralizing gods following a general pattern
of evolution from small (BSP) to big (MHG)—the presence or absence of high
gods independent of their moralizing status has little functional relationship with
social complexity, and instead appears largely contingent on history and geog-
raphy. In regions such as southern and eastern Asia, BSP in the form of karmic
religions (Buddhism and Hinduism) remains the dominant form of moralizing
gods, whereas in regions such as Europe and Africa moralizing Abrahamic MHGs
were commonly adopted or imposed without any intermediate evolution through
a BSP stage.
Doctrinal rituals. The modes of religiosity hypothesis focuses on two factors that
facilitate standardization of a body of beliefs and practices. First, high frequency
(for example, daily or weekly) collective rituals facilitate easy detection of devia-
tions from the orthodox canon. Second, religious hierarchy enables enforcement
of authorized belief and practice. Seshat codes five different types of rituals: the
most frequent, most widespread, largest scale, most euphoric and most dysphoric
rituals. For each ritual, frequency is coded as daily, weekly, monthly, seasonally,
yearly, generationally or once-in-a-lifetime. Seshat also encodes levels of religious
hierarchy. One represents no levels of religious hierarchy beyond the local priest
or shaman, whereas higher numbers represent multiple levels of hierarchy (for
example, senior priests or high druids).
Making inferences about prehistoric rituals requires using various measurable
archaeological proxies. Previous research has established that both frequent rituals
and multi-level religious hierarchies tend to co-occur with other features of doctri-
nal rituals (for example, low arousal)25,37–39. Not all of these features can always be
found in the archaeological record, so in this paper we use the appearance of either
religious hierarchy or frequent rituals as proxies for the appearance of doctrinal
rituals. Doctrinal rituals were thus coded as present if the most frequent ritual
occurred weekly or daily, or if there was evidence of multiple levels of religious
hierarchy.
Separate re-analyses were also conducted in which doctrinal rituals were defined
based only on ritual frequency and only on religious hierarchy. In both cases,
doctrinal rituals still preceded moralizing gods by an average of over 200 years,
although this difference only remained significant when using religious hierarchy
as a proxy for doctrinal ritual practices (religious hierarchy: mean = 991 years,
t = 2.4, d.f. = 11, P = 0.035; ritual frequency: mean = 210 years, t = 1.1, d.f. = 11,
P = 0.30).
Note that we coded only aspects of ritual practices and religion associated with
the official cult, and so the rituals of interest were not necessarily polity-wide but
could be largely or wholly restricted to elite groups.
Data collation. The process of data collection for the MHG, BSP and doctrinal
rituals variables involved matching each fully trained research assistant with one
or more experts (recognized authorities on the polity in question, typically holding
a relevant doctorate and occupying a faculty position in a university). Experts
provided guidance on how to delineate the temporal and geographical boundaries
of the polity, assembled an initial reading list and—where necessary—helped to
interpret some of the key historiographical debates associated with the variables.
Research assistants then populated the variables with data and presented these
data to the experts for review. The comments and suggestions made by the experts
were then implemented by the research assistants. The next stage required a sec-
ond team of fully trained research assistants to conduct a series of quality checks,
including vetting of the footnotes (which currently reference over 2,000 unique
sources) and the use of correct syntax for the machine-readable part of the data.
Finally, this checked dataset was offered to the experts for review. By contrast, the
data required for the social complexity variables frequently consisted of facts that
research assistants could procure with less supervision, allowing expert input to
occur at a later stage of the process. Data vetting in Seshat is a continuously ongo-
ing dynamic process that includes incorporation of disagreement among experts
within the project and input from external experts via our open-access interface.
There is room for reasonable disagreement about the most effective way of
gathering data about world history, particularly regarding the role of expert con-
tributors40,41. An alternative approach would be to have every single data point
signed off by a single recognized expert, perhaps even without requiring further
citations. We trialed such an expert-driven approach to data entry during initial
phases of our project but found it took too long to source experts and have them
enter the data required. Instead we found that faster progress could be made using
the approach described above and having multiple points at which the data were
examined and vetted. To this end, we have made not only all our data but also all
of the metadata and references supporting these data available for everyone to
examine and comment on. Rather than relying on the authority of a single expert
for each entry, Seshat involves regional experts to help to guide data collection
and assess the quality of our data and metadata as one of several complementary
components in our quality control approach, which also includes incorporating
disagreement among multiple experts.
Analyses. To ensure consistency and comparability in our analyses, we sampled
polities at 100-year intervals, sampling whichever polity happened to occupy
a given region at ad 100, ad 200, ad 300 and so on, and not including polities
that existed only between century boundaries8 (see Supplementary Information
for details and examples regarding the temporal sampling procedure). All
analyses were performed in R v.3.4.142. All P values reported are from two-tailed
analyses.
Quantifying social complexity. To create an overall measure of social complexity,
we took a previously published approach based on principal component analy-
sis (PCA)8 and applied it to the latest available data from Seshat. This method
aggregates the 51 social complexity variables (Extended Data Table 5) into nine
complexity characteristics and then analyses them using PCA.
PCA is a commonly used tool for dimension reduction—in this case we have
nine different aggregated variables that we want to reduce to a single variable that
best captures social complexity. However, we obtain the same conclusions even
without using PCA, regardless of which of the nine complexity characteristics we
choose as a proxy for social complexity (Extended Data Table 4).
As previously shown, these different complexity characteristics turn out to be
highly correlated and all load heavily onto a single principal component that cap-
tures 76% of the variance in the individual complexity characteristic variables.
Our approach uses multiple imputation43 to account for missing data, uncertainty
and expert disagreement by imputing data based on a range of possible values
and averaging the results over the course of 20 imputations. The results of this
approach have proven highly robust to a number of different modelling assump-
tions8 (Supplementary Information). Full details of this approach and justifica-
tions for selecting the social complexity variables can be found in a previously
published paper8. We previously carried out a number of robustness checks8,
including cross-validation analysis and bootstrap resampling to assess whether
our PCA methods were robust to spatio-temporal autocorrelation. Specifically,
k-fold cross-validation showed that our multiple imputation methods accurately
predicted complexity characteristic values when each geographical region was sys-
tematically removed from the analysis, and bootstrapping showed that removing
different geographical regions and time periods did not affect our PCA results8
(see Supplementary Information for full details).
Extensions of PCA (for example, generalized low-rank models44, spatio-tempo-
ral PCA45 and singular spectrum analysis46) may be worth considering in future
analyses as alternative methods of accommodating binary variables and spatio-
temporal autocorrelation. Note, however, that the subsequent regression analyses
performed in this paper explicitly control for spatial, temporal and phylogenetic
autocorrelation. More importantly, our current results consistently failed to support
the temporal sequence of the moralizing gods hypothesis across all geographical
regions (Extended Data Table 1 and Extended Data Fig. 1), and robustness analyses
using each of the nine complexity characteristics independently without perform-
ing PCA also confirmed our main findings (Extended Data Table 4). This confirms
that our primary finding that complex societies precede moralizing gods cannot
be an artefact of autocorrelation in our PCA methods.
Logistic regression. To examine the association between moralizing gods and social
complexity while controlling for non-independence in our data owing to spatio-
temporal autocorrelation and historical connections between cultures47, we fitted
a logistic regression model to the data. A detailed description of this model
has previously been published along with extensive validation of its robustness
when applied to Seshat data48. This approach stems from the field of nonlinear
dynamical systems, and is similar to the concept of Granger causality49,50 (which
is commonly used in economics), in that both use linear models with time-lagged
variables.
Our approach is similar to a previously published study7, except that we use
more fine-grained measures of geographical diffusion and linguistic similarity,
and also incorporate temporal information, as follows:
 δi, j 
Yi, t = a + ∑ bτ Yi, t − τ + c ∑ exp −  Yj, t − 1 + h ∑ wi, jYj, t − 1
 d 
τ i≠ j   i≠ j
+ ∑ gkXk, i, t − 1 + εi, t
k
in which Yi,t is the binary variable encoding the presence or absence of moralizing
gods in location i at time t. The time step Δt = 100 years. Starting from the first
term on the right-hand side, a is the regression constant (intercept). The next
term captures the influences of past history (autoregressive terms), with τ = 1,
2, … indexing time-lagged values of Y (as time is measured in centuries, Yi,t − 1
refers to the presence or absence of moralizing gods 100 years before t). The third
term represents potential effects resulting from geographical diffusion51,52. We use
a negative-exponential form to relate the distance between society i and society
j (δi,j) to the influence of j on i because—in contrast to a linear kernel—a nega-
tive exponential does not become negative at very long δi,j; it instead approaches

0 smoothly. We avoid the problem of endogeneity by using time-lagged Yj,t − 1.
Thus, the third term is a weighted average of the occurrence of moralizing gods
in the vicinity of society i at the previous time step, with weights falling off to 0
as distance from i increases. Parameter d measures how steeply the influence falls
with distance, and was set to d = 1,000 km after optimizing the Akaike information
criterion value using 200-km increments from 200 to 2,000 km (d = 200, 400, 600,
…, 2,000 km). Parameter c is a regression coefficient measuring the importance
of geographical diffusion. Detecting autocorrelations owing to shared cultural
history (next term) is done analogously, except w now represents the weight due
to linguistic similarity (set to 1 if societies i and j share the same language, 0.5 if
they are in the same linguistic genus, 0.25 if they are in the same linguistic family,
and 0 if they are in different linguistic families; linguistic genera and families were
taken from Glottolog53 and the World Atlas of Language Structures54). The rest of
the right-hand side represents effects of predictor variables Xk,i,t − 1 (time-lagged);
gk are regression coefficients and εi,t is the error term. This approach allows us to
investigate the effects of the predictor variable (social complexity, calculated above
via PCA), while controlling for serial autocorrelations, spatial diffusion and auto-
correlations due to the shared cultural history. The regression results are detailed
in Extended Data Table 2.
Comparison of before and after moralizing gods. To more directly examine the
direction of causality predicted by the moralizing gods hypothesis, we created a
time series of social complexity over time for all 12 regions for which social com-
plexity data were available both before and after the appearance of moralizing gods
(Extended Data Fig. 1). We then compared rates of change in social complexity
over time before and after moralizing gods using sliding time windows. First, we
compared rates of change using a 100-year window (that is, comparing the rate
for the 100 years before the appearance of moralizing gods with the rate for the
100 years after), then repeated this using a 200-year window, 300-year window
and so on, up to a maximum of between 700- and 3,900-year windows depending
on the region (different regions have different time depths of data available for
making these comparisons).
The region with the shallowest time depth was in Mali (±700 years before/after
the appearance of Islam around 1100 ad (400 ad–1800 ad)), whereas the region
with the deepest time depth was in Iran (±3,900 years before/after evidence of the
moralizing Mesopotamian sun god Shamash around 2200 bc (6100 bc–1700 ad)).
If we used all available data (up to ±3,900 years), we risked weighting the analyses
too heavily towards regions such as Iran with deep time depths, whereas using
only a consistent upper limit of a maximum of ±700 years risks throwing away
too much data. As a compromise, we conducted analyses using an intermediate
upper limit of a maximum of ±2,000 years (Fig. 2b), but also repeated the analy-
ses using extreme upper limits of ±700 years and ±3,900 years (see ‘Robustness
analyses’). All of these choices produced qualitatively identical results (Extended
Data Table 4).
Note that these analyses do not attempt to construct a single average rate of
change before moralizing gods, a single average rate of change after moralizing
gods, and compare these average rates. We cannot assume such a constant rate of
change—and, indeed, Fig. 2a clearly shows that rates of change are not constant.
Instead, these analyses calculate a difference value for each time window (for exam-
ple, subtracting the rate of change for the 100-year period before moralizing gods
from the rate of change for the 100-year period after, and then doing the same
for a ±200-year period, and so on). The key prediction of the moralizing gods
hypothesis is that these difference values should tend to be positive (that is, for a
given time window, the rate of change after moralizing gods should be greater than
the rate before). However, Fig. 2b demonstrates that—in fact—the distribution of
difference values was significantly negative (paired t-tests, P < 10−9).
Robustness analyses. To explore the robustness of our results to modelling assump-
tions, we ran the following robustness analyses.
To ensure that the analyses are not affected by the fact that religious hierarchy
is included as one of the social complexity variables in addition to being one of
the variables used to define doctrinal mode, we reran the analyses after removing
the religious hierarchy variable from the social complexity variables. We chose to
do this for robustness analyses rather than the primary analysis in order to use the
same 51 social complexity variables used in our previously published studies8,48
for consistency.
To ensure that the observed plateauing of social complexity was not simply an
artefact of a ceiling effect wherein polities ‘max out’ certain variables, we reran
the analyses twice after splitting the social complexity variables in two subsets.
The scale subset contained only the subset of seven social complexity variables
for which there was no theoretical maximum value (from the categories polity
population, polity territory, capital population and hierarchy). The non-scale sub-
set contained the remaining 44 social complexity variables for which there was a
theoretical maximum that could be attained once all our variables were present
in a society (from the categories government, money, infrastructure, information
systems and texts).
Letter RESEARCH
To examine whether our results were affected by the definition of moralizing
gods, we reran the analyses limiting the definition of moralizing gods exclusively
to MHG, rather than the more inclusive definition of BSP used in the primary
analysis.
Our primary analysis treated moralizing gods as being present from the begin-
ning of the polity in which they appeared. To ensure that our analyses were not
affected by dating uncertainty, we reran the analyses randomly resampling to treat
moralizing gods as appearing at some point from within the full date range of this
polity (for example, 2900–2700 bc for Egypt).
Our primary analysis used time windows of up to 2,000 years before and
after the appearance of moralizing gods, because 2,000 years was intermediate
between the maximum time window for the region with the shallowest time depth
(±700 years for Mali) and the deepest time depth (±3,900 years for Iran). To
examine whether our results were affected by the depth of the time window used,
we reran analyses using consistent time windows for each region of up to 700 years
before/after moralizing gods (because ± 700 was the maximum time window pos-
sible for Mali), and also using the full time window available for each region (that
is, as wide as 3,900 years for Iran).
To ensure that our results were not affected by possible autocorrelation in our
use of PCA to extract a measurement of social complexity, we reran the analysis
nine times using each of the nine individual complexity characteristics as a measure
of social complexity without performing any PCA.
All of these robustness analyses (16 in total) produced qualitatively identical
results in which the rate of increase of social complexity was significantly greater
before the appearance of moralizing gods than afterwards (more than double in all
cases; Extended Data Table 4), which confirms that our primary conclusion that
complex societies precede moralizing gods is highly robust.
Reporting summary. Further information on research design is available in
the Nature Research Reporting Summary linked to this paper.
Code availability. Source code is available online at http://github.com/pesavage/
moralizing-gods.
Data availability
The full machine-readable dataset is available as Supplementary Data 1, and at
http://seshatdatabank.info/datasets. Full coding data with detailed explanations
and references are available at http://seshatdatabank.info/data, and are summarized
in Supplementary Table 2. The data include the coded levels of uncertainty and
disagreement, the textual explanations and the references for each of the variables
for all polities used in our analysis. These webpages also make it possible to com-
ment on each of our data points and suggest additions or corrections and thus
provide an up-to-date and dynamic dataset that undergoes continual improvement
by members of the Seshat team and external scholars. To maximize transparency,
we have tied each cluster of variables to the names of the research assistants who
gathered the data, and to the names of the experts who reviewed the data.
31. Swanson, G. E. The Birth of the Gods: The Origin of Primitive Beliefs (Michigan
Univ. Press, Ann Arbor, 1960).
32. Divale, W. Pre-Coded Variables for the Standard Cross-Cultural Sample Vols 1 and
2 (York College, New York, 2000).
33. Johnson, D. God is Watching You: How the Fear of God Makes Us Human (Oxford
Univ. Press, New York, 2016).
34. Norenzayan, A. Big Gods: How Religion Transformed Cooperation and Conflict
(Princeton Univ. Press, Princeton, 2013).
35. Haidt, J. & Graham, J. When morality opposes justice: conservatives have
moral intuitions that liberals may not recognize. Soc. Justice Res. 20, 98–116
(2007).
36. Curry, O. S. in The Evolution of Morality (eds Shackelford, T. K. & Hansen, R. D.)
27–51 (Springer, Cham, 2016).
37. Whitehouse, H. & Lanman, J. The ties that bind us: ritual, fusion, and
identification. Curr. Anthropol. 55, 674–695 (2014).
38. Whitehouse, H. & Martin, L. H. Theorizing Religions Past: Archaeology, History, and
Cognition (AltaMira, Blue Ridge Summit, 2004).
39. Atkinson, Q. D. & Whitehouse, H. The cultural morphospace of ritual form:
examining modes of religiosity cross-culturally. Evol. Hum. Behav. 32, 50–62
(2011).
40. Slingerland, E. & Sullivan, B. Durkheim with data: the Database of Religious
History. J. Am. Acad. Relig. 85, 312–347 (2017).
41. Wade, L. Birth of the moralizing gods. Science 349, 918–922 (2015).
42. R Core Team. R: A Language and Environment for Statistical Computing.
https://www.r-project.org/ (2015).
43. Rubin, D. B. Multiple Imputation for Nonresponse in Surveys (Wiley, New York,
1987).
44. Udell, M., Horn, C., Zadeh, R. & Boyd, S. Generalized low rank models. Found.
Trends Mach. Learn. 9, 1–118 (2016).
45. Stahlschmidt, S., Härdle, W. K. & Thome, H. An application of principal
component analysis on multivariate time-stationary spatio-temporal data. Spat.
Econ. Anal. 10, 160–180 (2015).
46. Hassani, H. Singular spectrum analysis: methodology and comparison. J. Data
Sci. 5, 239–257 (2007).
 RESEARCH Letter

47. Mace, R. & Pagel, M. The comparative method in anthropology. Curr. Anthropol.
35, 549–564 (1994).
48. Turchin, P. Fitting dynamical regression models to Seshat data. Cliodynamics 9,
25–58 (2018).
49. Granger, C. W. J. Investigating causal relations by econometric
models and cross-spectral methods. Econometrica 37, 424–438
(1969).
50. Kleinberg, S. Causality, Probability, and Time (Cambridge Univ. Press, Cambridge,
2013).
51. Eff, E. A. & Dow, M. M. How to deal with missing data and Galton’s problem in
cross-cultural survey research: a primer for R. Struct. Dyn. 3, 1–29 (2009).
52. Eff, E. A. & Routon, P. W. Farming and fighting: an empirical analysis of the
ecological-evolutionary theory of the incidence of warfare. Struct. Dyn. 5, 1–33
(2012).
53. Hammarström, H., Forkel, R., Haspelmath, M. & Nordhoff, S. Glottolog version
2.3 http://glottolog.org (2014).
54. Dryer, M. S. & Haspelmath, M. The World Atlas of Language Structures online
http://wals.info (Max Planck Institute for Evolutionary Anthropology, 2013).
 Letter RESEARCH

Extended Data Fig. 1 | Social complexity time series for individual bands represent the period in which the first evidence of moralizing gods
regions. The 12 regions for which social complexity data are available both (red) and doctrinal rituals (blue) appeared. Grey shading represents 95%
before and after the appearance of moralizing gods are shown. Vertical confidence intervals based on a PCA using multiple imputation8.
RESEARCH Letter
Extended Data Fig. 2 | Full time series showing mean social complexity
over time before and after the appearance of moralizing gods. n = 12
regions with data before and after the appearance of moralizing gods.
Social complexity has been scaled so that the society with the highest
social complexity (Qing Dynasty, China, around ad 1900) has a value
of 1 and the society with the lowest social complexity (Early Woodland,
Illinois, USA, around 400 bc) has a value of 0. Vertical bands represent
the period in which moralizing gods and doctrinal rituals first appeared.
All errors represent 95% confidence intervals, with the exception of the
vertical bar for moralizing gods, which represents the mean duration
of the polity in which moralizing gods appeared (because times are
normalized to the time of first evidence of moralizing gods, and there is
thus no variance in this parameter). Lack of confidence intervals indicates
data from only a single region. This figure is identical to Fig. 2a, except
that it also includes all available data before and after moralizing gods,
rather than being restricted to a window of 2,000 years before and after.

Extended Data Fig. 3 | Social complexity before and after the
appearance of MHG. This is a version of Fig. 2 in which analyses are
restricted to only MHG, rather than the broader definition of moralizing
gods used in Fig. 2 and elsewhere (which includes BSP as well as MHG).
a, Time series showing the mean social complexity over time for
2,000 years before and after the appearance of MHG. n = 10 regions with
social complexity data for before and after moralizing high gods. Social
complexity has been scaled so that the society with the highest social
complexity (Qing Dynasty, China, around ad 1900) has a value of 1 and
the society with the lowest social complexity (Early Woodland, Illinois,
Letter RESEARCH
USA, around 400 bc) has a value of 0. Vertical bands represent the period
in which MHG and doctrinal rituals first appeared. All errors represent
95% confidence intervals, with the exception of the vertical bar for MHG,
which represents the mean duration of the polity in which MHG appeared
(because times are normalized to the time of first evidence of MHG
and there is therefore no variance in this parameter). b, Histogram of
differences in rates of change in social complexity after minus before
the appearance of MHG (n = 158 time windows from the 10 regions).
The y axis represents the number of time windows out of 158.
 RESEARCH Letter

Extended Data Table 1 | Timing and rates of change in social complexity before and after the earliest precolonial evidence of moralizing
gods

For locations without precolonial concepts of moralizing gods, the polity represents the latest polity analysed. See Supplementary Table 2 and http://seshatdatabank.info/data for details and
references. Rates of change before and after moralizing gods were compared using paired t-tests on up to 20 time windows (100–2,000 years before and after the appearance of moralizing gods)
for all 12 regions with social complexity data available both before and after the appearance of moralizing gods. Negative t-values represent higher rates of change before moralizing gods. *P < 0.05;
**P < 0.01; ***P < 0.001.
 Letter RESEARCH

Extended Data Table 2 | Logistic regression results predicting moralizing gods

The model includes parameters for social complexity and for geographical, temporal and cultural relationships, ordered by absolute z-value (see Methods for details). SE, standard error. Pr, probability.
 RESEARCH Letter

Extended Data Table 3 | Analyses with doctrinal rituals instead of moralizing gods as the dependent variable

See Supplementary Tables 3–5 for full regression results.

 Letter RESEARCH

Extended Data Table 4 | Robustness analyses modifying modelling assumptions of the analyses

See Methods for details, and see Extended Data Table 2 and Supplementary Tables 6–18 for full regression results. NA, not applicable.
 RESEARCH Letter

Extended Data Table 5 | List of the 51 social complexity variables analysed

See a previous study8 and http://seshatdatabank.info/methods/codebook/ for full definitions and selection rationale.
 nature research | reporting summary
Corresponding author(s): Patrick E. Savage
Last updated by author(s): Feb 5, 2019

Reporting Summary
Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency
in reporting. For further information on Nature Research policies, see Authors & Referees and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement
A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly
The statistical test(s) used AND whether they are one- or two-sided
Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested

A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons
A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient)
AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)

For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted
Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings
For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes
Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated
Our web collection on statistics for biologists contains articles on many of the points above.

Software and code

Policy information about availability of computer code
Data collection All data was collected via, and is available open-access at: http://seshatdatabank.info/data/

Data analysis All analyses were performed in R V3.4.2 and R Studio V 1.1.383. Source code is available online at http://github.com/pesavage/
moralizing-gods
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors/reviewers.
We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data
All manuscripts must include a data availability statement. This statement should provide the following information, where applicable:
- Accession codes, unique identifiers, or web links for publicly available datasets
- A list of figures that have associated raw data
- A description of any restrictions on data availability

The full machine-readable dataset is available as Supplementary Dataset 1, and at http://seshatdatabank.info/datasets. Full codings with detailed explanations and
October 2018

references are available at http://seshatdatabank.info/data, and are summarized in Extended Data Table 2. The data include the coded levels of uncertainty and
disagreement, the textual explanations, and the references for each of the variables for all polities used in our analysis.

1
 nature research | reporting summary
Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.
Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design

All studies must disclose on these points even when the disclosure is negative.
Sample size The global sample of 30 regions was selected and published (Turchin et al., 2015) before beginning data collection. The number of polities per
region depended on the availability of historical and archaeological data, and was determined by the relevant expert historian/
archaeologist(s) for each region.

Data exclusions The time period of interest was limited to between the beginning of the Neolithic to the beginning of Colonial/Industrial periods. No data
falling within these periods was excluded.

Replication No experiments were performed. Our primary results were replicated in all robustness analyses (see Methods).

Randomization No randomization was performed.

Blinding Historical data were coded by research assistants and regional experts who could not be blinded to the society they were coding. However,
they were blind to the specific hypotheses being tested.

Reporting for specific materials, systems and methods

We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material,
system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Materials & experimental systems Methods

n/a Involved in the study n/a Involved in the study
Antibodies ChIP-seq
Eukaryotic cell lines Flow cytometry
Palaeontology MRI-based neuroimaging
Animals and other organisms
Human research participants
Clinical data

October 2018